In dynamic multi-pedestrian environments, while a service robot navigates only relying on its own sensors with the first-person perspective, both the uncertainties of robot localization and the estimation of people's states are increased, which hinder the navigation decision of a service robot. To solve this problem, a local collision avoidance method based on the optimal reciprocal collision avoidance (ORCA) is proposed. In this method, the states of multiple pedestrians are estimated by a variant of particle-PHD filter, i.e. NP-PHDF for multi-target tracking. NP-PHDF is a combination of Kalman particle filter and PHD filter and has their advantages. So it can not only track time-varying number of targets with sudden motion changes such as abrupt acceleration/deceleration or steep turn, but also resist block among pedestrians. Meanwhile, similar as robot localization which uses particle filter, the uncertainties of estimation for pedestrians can be represented by the distribution of particles. To reduce the uncertainties, an encircling-particles method is proposed to refine the true states of robot and pedestrians from the probabilistic particle distribution. The effectiveness of the proposed technique is demonstrated through experiments in real environments.